When I saw How to Lie with Statistics in the free books boxes I jut had to grab it. This is the grand-daddy of line of books about bias in statistics which lead to such descendants as John Allen Paulos’ A Mathematician Reads the Newspaper. Turns out it’s a short read that uses humor to impress the subject on the population rather than stridency.
The line about statistics that most people quote is from Benjamin Disraeli, via Mark Twain: There are three kinds of lies: lies, damned lies, and statistics.
How to Lie with Statistics explores the ways that statistics can commonly be used to create misinformation. But something that Darrell Huff is careful to make clear is that statistics themselves aren’t bad and they do not lie any more than any other set of facts. The lying comes from poor knowledge of statistics, common psychological pitfalls, and unintentional bias and intentional misuse by the proponents of certain arguments.
Most of the topics covered are ones with which I am already familiar, having read the works of Paulos. Credit properly ought to go to Huff since he wrote this much earlier than Paulos did. Still I read the younger popular mathematician’s work first so I am not going to rehash the topics in this review. He covers such stuff as sampling bias, misuse of the term average
, not knowing the significance and ranges of figures, misleading graphs, and more.
One topic that either I glossed over in my prior reading or the authors glossed over is that of changing the subject
as Huff puts it. It’s where the proponent of an argument quotes one statistic that has little to do with the ultimate argument he/she is making, but seems relevant. The best example (according to me) from the book is the claim that it is safer to drive at 7 a.m. than at 7 p.m., because four times as many accidents occur in the evening than in the morning. But it’s sleight of statistics, because the relevant denominator
is not time of day but instead number of miles driven. Approximately ¼ as many people drive in the morning, cutting accidents by a similar amount. If we start driving in the morning to avoid accidents based on this perfectly true but misleading statistic, accidents in the morning will rise proportionally.
One other thing I liked is his concluding chapter is his method for talking back
to statistics to see how relevant they actually are. I’m a little skeptical that his five questions would really be helpful. It’s not that they are wrong, but using them to actually find misinformation will require statistical expertise and familiarity that the public just doesn’t have. His questions to ask are:
- Who says so?
- How does he know?
- What’s missing?
- Did somebody change the subject?
- Does it make sense?
Lord knows people’s sense
includes lots of things that just aren’t true. Hello astrology! It’s a good set of questions to ask if one has that familiarity, not so helpful if one doesn’t. But it doesn’t really hurt to have the questions to guide readers and at least it makes a stab at a solution.
Title: How to lie with statistics
Author: Darrell Huff
Illustrator: Irving Geis
Imprint / publisher: W. W. Norton & Company
Format: Paperback
Length: 142 p.
Publication date: 1954
ISBN-10: 0-393-09426-X
Subject: Statistics
LC classification: HA29 .H82
